You send an address, a list of addresses, or a territory with your constraints. We ask the questions your brief left open, plan the steps, gather the data, and write the work into the room: the figures, the sources behind them, and which numbers are assumptions. Most jobs land within the hour.
Step 1. You send the subject
An address, a list of addresses, or a territory, plus the constraints that matter: budget, size, target yield. Most jobs need nothing else. Two do: Pitch Deck and OM need a trailing-twelve operating statement, and will not start without one.
Most people send very little: the median client message is 43 characters, a task stated under time pressure rather than a specification. That is enough.
Step 2. We ask you the rest
The conversation is an interview, not a form. Expect a few questions after your first message: the asset type, the size, what you are trying to decide. Most people answer in a line or two. Without the asset type or the size the job does not start well, and asking is faster than guessing.
You can upload files, and most working conversations include one. Reading inside them is weaker: an arbitrary spreadsheet, a shared drive file or a link to a listing page often will not be parsed. If a number matters, type it into the message as well.
Step 3. The system plans the work and runs it
The job is broken into a chain of steps. Data is gathered from registry, transaction and open sources. The result is written into the room where you started, with the figures and where each came from.
Measured on finished runs in the 30 days to 18 October 2026, the median time from request to delivered result was about 58 minutes for Underwriting, 54 for Scout, 45 for a Value Opinion, 28 for BOV at Scale and for Feasibility, 19 for Comps and 16 for Market Rent. The 90th percentile is the number to plan around if you have a meeting: about 95 minutes for Underwriting, 95 for Scout, 72 for Feasibility. The tails are long; if you need a commitment rather than an estimate, ask and you will be given the slow end.
Step 4. What comes back, and where it comes back
A written answer, in the room. For several products a workbook you can download comes with it, and a report can be sent to someone else by link.
We counted the runs that were created in the thirty days to 2 October 2026 and delivered: 623. Of those 623, 211 came with an Excel workbook: 84 of 86 Market Rent runs, 45 of 51 Feasibility runs, 32 of 55 Value Opinion runs and 18 of 39 Underwriting runs. BOV at Scale, Comps and Scout returned almost none (0 of 129, 1 of 75 and 1 of 64): their result is the answer and the tables in the room.
A report can be shared by link, and it is rarely done: ten share links exist in the whole history of the system, seven of them made in those thirty days, and three have been revoked.
Where we prepared a deal package, a letter went out with a link to a page that holds its files: ten such letters in the same thirty days.
We print no figure for slide decks here. How many runs return one has not been measured the same way, and we would rather leave the number out than guess it.
Step 5. What gets checked before you see it
This is the shape of a finished underwriting deliverable: a 416-unit apartment property in Tampa that our own analysts modelled from property records, not from a client file. The workbook has six tabs, and the last one classifies every number in it as sourced, assumed, or computed; nothing else is allowed on that tab. The debt tab runs both a loan-to-value test and a debt-service-coverage test every time and states which of the two bound the loan, because that decides how much equity the deal needs.
The first draft of that model was wrong. It returned a 53% levered internal rate of return, a 6.36x equity multiple, and a debt service coverage ratio of 2.98x, because it had priced the building at a tax figure instead of at what its own income supports. Price and income sat on different bases, and the arithmetic reported the gap as return. After the fix, two independent recalculations of the corrected model agreed to the cent on every output, and put the internal rate of return at 8.66% and 8.7%.
Four checks were then built into the file itself rather than a cover note: a going-in yield outside roughly 3.5–9%, an exit capitalisation rate priced tighter than the going-in rate, a coverage ratio above 2x, and an operating expense ratio below 40% for Florida product. Each of those had fired on the bad draft.
That example carries its own limit, stated on the tab where it matters: no rent roll and no trailing-twelve operating statement for that building was ever seen, so every rent, vacancy rate, expense line and both capitalisation rates in it are assumptions. A model is a checked structure until your own numbers go into it.
What the output is
Our figure is an opinion of value. It is not an appraisal and we hold no appraisal licence. Where fewer than three comparable sales are available, the output is a range rather than a point, and says so.
What we need from you, and what happens if you cannot provide it
Your own documents. A rent roll or a lease expiry schedule cannot be built from property records. If you cannot share them, we will run everything else and mark those lines as assumptions for you to overwrite, rather than invent them.
A property inside our coverage. As of 2026-09-25 connected sources sit under 49 jurisdiction codes (48 states and DC) and two states have none at all: Kansas and South Dakota. South Carolina was on that list on 2026-09-11 and is not on it now. A territory search where we hold nothing still does nothing, and a connected source is not the same as a deep one. Texas is covered, but does not disclose transaction prices; the state's rule, which no vendor fills, so Texas sale prices are not available here. The system will not reliably warn you that a subject is outside coverage, so ask before you run.
A question we are allowed to answer. The analysis does not apply your jurisdiction's rules on tax, tenure or measurement: that pack does not reach the model, in any country. We do not give legal or tax advice. We also cannot return current advertisement listings with links; we work from registry, transaction and open data.
What happens to what you send
Your material is isolated to your own account and nothing is pooled with another firm's. Accounts are individual: there is no shared firm workspace and no administrator view, so work is passed on by sending it. Runs execute through third-party model providers, six of which hold credentials in our production system. We have no training pipeline and do not train on client material; what each provider does under its own API terms is a separate question, and we would rather come back with the specifics than give a blanket assurance.
If something does not arrive
If you typed something and nothing happened, tell us. That is a known failure at our door rather than a mistake at yours, and the first thing we check is whether a run ever started.